Bayesian Systems for Optimizing Treatment Protocols in Oncology

The development of new pharmacodynamic biomarkers has greatly increased the information content of clinical trials, and made possible the construction of pharmacokinetic/pharmacodynamic (PK-PD) models. A population PK-PD model, in conjunction with a disease model, can then be used to simulate clinical trials in silico. In the case of oncology, the disease model must describe the cancer cell cycle, and such aspects of tumour biology as growth, invasion, metastasis, angiogenesis, and interactions of tumour cells with the immune response. Such models of tumour growth and its response to drug treatment can be used in conjunction with databases of drug PK and PD parameters and with databases of biomarkers, to ask the question: for a tumour with a given biomarker expression profile, what treatment protocol is likely to be most effective? Various treatment options can then be modelled, and the one predicted to be most active selected for clinical evaluation. Since, in practice, many of the tumour growth parameters are still unknown, a Bayesian approach is required: prior assumptions are made, based upon preclinical data and historical precedent. The course of treatment, based upon these prior assumptions, is predicted, compared with the clinical outcome, and the difference fed back to drive model adjustments. The assumptions of the model are thus progressively refined - the system learns from experience. A Bayesian model can be used to devise optimal control strategies for chronic disease. A frequent cause of cancer treatment failure is the rapid development of acquired drug resistance. In silico clinical trials that incorporate the techniques of evolutionary dynamics can be used to predict the incidence of drug resistance, including multi-drug resistance, so that emergence of resistant clones can be minimized and delayed. In due course, the actual clinical outcome can then be compared with the predictions. We review the literature on the use of Bayesian systems in oncology, and discuss their application to development strategies for new drugs, and to developing personalised medicine approaches in oncology.

Keywords: Bayesian networks, Cyclotherapy, Evolutionary dynamics, Expert systems for chemotherapy, Immune system modelling, In silico clinical trials, Optimal control, Pharmacokinetic-pharmacodynamic modelling, Systems pharmacology, Virtual interactive patient, Virtual tumour.